Virtual Insider

Blog archive

vSphere 5 Top 10: Storage DRS at #1

The countdown concludes with the top 2 finalist features of vSphere 5. At number 1, I have Storage DRS, a fantastic new feature that load balances virtual machines across the different datastores to optimize performance.

Up until vSphere 5, DRS was a cluster function limited to load balancing compute resources (CPU and memory), and that is a great feature because as VMs start to consume a lot of physical resources on a particular ESXi host, DRS would instruct vMotion to migrate VMs to a more suitable host that can better satisfy their resource needs. In the beginning this was enough--we were just virtualizing some servers and some of us were testing and using it in development environments.

Now that virtualization has gone mainstream in production and the consolidation ratios have increased, it has become increasingly more difficult to load balance VMs across datastores. Moreover, those of us that were successful in an initial deployment of properly load balancing VMs across datastores recognized quickly that maintaining this load balancing was not easy.

Storage DRS now increases the functionality of our clusters by extending VM load balancing across the different datastore leveraging Storage vMotion. As a result of this immensely helpful feature, it's number 1 on my vSphere 5 list.

By now you might have guessed the number 2 feature is the completely re-architected High Availability. Cosmetically HA looks and feels the same to the virtualization admin, but there have been significant changes under the hood.

First, HA no longer requires DNS to properly function. In earlier versions, this was a gotcha that many neglected and it indeed caused issues. That being said, my favorite new enhancement to HA has got to be the fact that it no longer relies on just the management network to determine if a host has failed. The new HA in vSphere 5 will check the management network. If it detects a failure, it will then use the storage connection as a secondary way of verifying if the host really failed. If it cannot contact the host using either method, only then will it deem the host as failed and HA will kick in.

There are other changes to the way HA works, like the fact there no longer is a primary/secondary relationship between the hosts in the cluster. Instead, there's a single master server which is determined via an election process. All other hosts in the cluster are slave hosts. Elections occur at different intervals, but it occurs primarily when a new host is added or if a host is rebooted. There are other things that can also trigger an election but these two events are in the forefront.

Have you been testing SDRS and the new HA? If so, do you agree that these are the two best features of vSphere 5?

Posted by Elias Khnaser on 09/08/2011 at 12:49 PM


Subscribe on YouTube